Trust-native AI outreach

AI does the outreach.
A human owns every send.

reach-os ranks creators worth contacting and shows its working, drafts in your voice from your own data, and runs the reply loop on its own — then stops and hands the thread to you the moment money or anything delicate comes up.

No fabricated metrics. No auto-negotiation. Runs end to end with zero secrets for offline development.

Draft ready · awaiting your approval

Hi Maya — your teardown of the Northline pack is the most careful review we've seen. We'd like to send you the new one before it ships.

fit 87 · shownno claims made
Money not mentioned — loop may continue
Where the world's creators already are
InstagramTikTokYouTubeTwitchXInstagramTikTokYouTubeTwitchXInstagramTikTokYouTubeTwitchX
The opening

Creator inboxes are filling with AI that
negotiates on nobody's authority.

A wave of tools now promises to discover, personalise and settle rates entirely on their own. That is a rush, not a moat. As the spam lands, the thing worth paying for becomes the opposite: a system that never fabricates a metric, shows why it recommended someone, and keeps a person on every send and every dollar.

How it works

From cold list to warm reply, on autopilot.

1

Find creators worth contacting

Discovery is web-grounded — it surfaces real, findable accounts and never invents a handle or a follower count. Every candidate arrives pre-ranked by the same fit score the rest of the pipeline uses, and marked unverified until you say otherwise.

2

Draft in your voice

The model writes a first contact from your own data. Creator-supplied text is encoded into an untrusted region before it ever reaches a prompt, so nothing in a bio can rewrite your instructions. You approve the send.

3

Run the reply loop

Inbound replies are classified for intent, tone, and confidence, and answered turn after turn — with a content-hashed record of what the model drafted and which human released it.

4

Stop at the human line

Price, contract terms, an opt-out, a cold turn, low confidence, or three automated turns — any one of them ends the automation and hands you a thread that is already warm.

The loop

Not a bot that never stops.
A rep that knows when to.

Every reply is read for intent, tone, and confidence. A soft yes or a simple question? reach-os keeps it moving on its own. The instant it gets delicate, a human takes over — warm, in context, never a beat too late.

  • Someone opts out → it stops. Full stop.
  • Price or contract terms come up → straight to a human.
  • Sentiment turns cold → your team takes over.
  • It's not confident → it asks you, not the creator.
  • After 3 automated turns → a person always steps in.
@lens.mara
Instagram · replied 4 min ago
reach-os
Hey! Love the brand — what did you have in mind? 👀
Reply classified · interested · positive · confidence 0.86
So glad! We'd love to send our new set for you to feature — your style, your call. Want the details?
Sounds good. What's the budget, though?
Money raised → warm handoff to your team
Twenty seconds, no narration: a draft is written, a person reads it, and a person releases it. That last step is the product.
Everything in one place

The whole outreach machine, minus the busywork.

Explainable fit scoring

Rank creators on niche, audience, engagement, platform — and geography when you target it. Every point is attributable; competitors return a match percentage and no reason for it.

Prompt-injection defence

Stored creator text is never mutated; a separate prompt-safe copy is what enters the model. Instruction-override and role-marker attempts are flagged to the send gate, not silently stripped.

The auto-conversation loop

Multi-turn replies that keep a lead warm — and a send gate that knows exactly which turns it is not allowed to take on its own.

Isolation the database enforces

Tenant separation is a row-level-security policy on org_id, not application logic that a missing WHERE clause can defeat. It is proven by a standing check suite.

Attribution you can measure

First-party tracking links, per-creator promo codes, UTM capture, and a secured conversion postback — so reported revenue is measured rather than estimated.

Authenticity with its reasons

Fake-follower and brand-safety signals arrive as named findings with explicit confidence and a source. A black-box authenticity percentage is exactly what we refuse to emit.

What's actually built

A tool that claims it never overstates has to start here.

Every line below matches the capability table in the repository. Where something is half-built or deliberately switched off, it says so.

Shipped

  • Explainable fit scoring, every point attributable
  • Web-grounded creator discovery, wired end to end
  • Audience crawler — an owned index from public signals, unknowns disclosed with a source
  • Explainable authenticity and brand-safety scoring
  • Content-fit scoring against your brand voice
  • The inbound auto-conversation loop and its send gate
  • Human-approved sending and first-contact delivery
  • Measured attribution — tracking links, promo codes, UTM, secured postback
  • Content-hashed audit trail: what the model drafted, who released it
  • Prompt-injection defence wired into the send gate
  • DB-enforced tenant isolation, org to brand to campaign
  • Briefs, contracts, ROI reporting and CSV export

Not yet, and why

Creator payouts are mock-provider only

Payouts are human-approved and run against a mock provider. The live transfer path deliberately throws until it is wired. No real money moves through reach-os today.

Billing is complete in test mode

The Stripe integration — webhook write path, idempotency, entitlements, customer portal, creation gating — is finished and exercised against test keys. Turning on live keys is an operator gate, not a shipped state.

Contracts do not e-sign yet

Briefs become contracts with a tracked status. Integration with a signing provider is still open work.

No licensed creator-data layer

Audience figures come from public web signals, so structured age, gender and geography splits — and real-reach and overlap — still need a richer source than we have.

No UGC rights or ad-whitelisting

Spark and Partnership Ads permissions are not built. There is no code for it yet.

Agency infrastructure is partial

Roles and multi-brand exist in the schema; the seats-and-invites interface and white-label reporting are not finished.

12/12
isolation checks your data passes
3
auto-replies before a human steps in
5
platforms your creators live on
100%
of messages you own & can edit
Pricing

Start free. Upgrade when it's paying for itself.

Money changes what you can do — never how hard the tool works for you. Annual plans get 2 months free.

Free

For your first outreach push.

$0 forever
  • 25 prospects
  • AI-drafted first contact + follow-ups
  • Fit scoring & guardrails
  • Manual reply handling
  • 1 brand workspace
Start free
Most popular

Pro

When outreach becomes a system.

$95 / month
  • 1,000 prospects
  • The auto-conversation loop
  • Live AI generation
  • Send through your own domain
  • ROI reports & CSV export
  • Month-to-month, cancel anytime
Start Pro

Growth

For teams and agencies running volume.

$245 / month
  • Everything in Pro
  • 3× the AI conversation budget
  • Multiple brand workspaces
  • White-label client reporting
  • Priority processing & support
Start Growth

Done-for-you

We run the whole outreach motion for you inside reach-os — sourcing, vetting, conversations, contracts, payouts. Founding-client rates for the first five brands or agencies.

Talk to a human

Leverage, with your name on it.

Start free, in minutes. Every message stays yours to read, edit, and release.